AGENT SKILL

auto-research

Research uncertain questions with an explicit, user-approved web search or ChatGPT consultation, then present options and wait for implementation approval.

License 许可
MIT
最近更新
2026-10-22
Stars
★ 1
Forks
0

01 前言 Front Matter

name
auto-research
risk
critical
tags
[ "research", "chatgpt", "playwright", "browser-automation", "decision-support", "chinese" ]
tools
[ "claude", "playwright" ]
author
zyu51
source
self
license
MIT
category
automation
date_added
2026-07-09
description
Research uncertain questions with an explicit, user-approved web search or ChatGPT consultation, then present options and wait for implementation approval.
source_type
self

03 SKILL.md 正文

Auto-Research Skill

Overview

When implementing tasks, Claude Code can encounter uncertainties — design choices, algorithm details, API usage, or best practices. This skill provides an explicit-consent research path, presents findings, and waits for user approval before writing code.

The skill supports web research and an optional ChatGPT consultation. It never sends conversation context, files, browser state, or credentials to a third party without the user's explicit approval of the exact, redacted text.

When to Use This Skill

  • User asks a question where multiple valid approaches exist
  • Claude is uncertain about algorithm details or API usage
  • Design/architecture choices need comparison
  • The user explicitly asks to search the web or consult ChatGPT and approves the proposed query

How It Works

Step 1: Propose the research boundary — State the source to use, the exact query or redacted prompt, whether any local/workspace text would leave the machine, and the likely cost. Wait for the user to approve that exact boundary.

Step 2: Research — After approval, use web search or a browser session the user has explicitly selected and authorized. Use a pinned, user-configured browser automation connector; do not install packages automatically, use @latest, or access browser cookies, other tabs, saved passwords, or sessions.

Step 3: Present — Distill findings into concise options with sources, presented to the user.

Step 4: Await Approval — Do NOT write code until the user says "go ahead" or picks an option.

Step 5: Implement — Once approved, execute with confidence.

Explicit ChatGPT Consultation

Do not treat ?, ??, or another shorthand as consent. First propose a minimal prompt, for example: 请评估这个已脱敏的方案的正确性、完整性和可改进之处:<text>. Explicitly identify every piece of text that would be sent. Only after the user confirms the exact prompt may you open the selected ChatGPT session, submit that prompt, and present the response. Do not include conversation history by default.

Redact secrets, personal data, proprietary code, customer data, and internal URLs before proposing the prompt. If safe redaction is not possible, do not submit it.

Browser Automation Boundary

If browser automation is necessary, the user must separately authorize the selected browser profile and connector version. Restrict the session to the consultation tab. Do not inspect, reuse, export, or rely on cookies from other tabs or profiles.

Examples

Example 1: Design Question with GPT

User: PyTorch 中自定义 ADMM 优化器怎么设计?
Claude: 我可以搜索公开资料,或将以下已脱敏问题发给 ChatGPT:
        “如何设计 PyTorch 自定义 ADMM 优化器?请比较可行模式。”
        不会发送工作区文件或对话历史。是否允许?
User: 允许发送这段文字
Claude: [Opens only the authorized consultation tab, submits the approved prompt]
Claude: GPT suggests approach A with these pros/cons. Proceed?
User: 行
Claude: [Implements code]

Example 2: Web Search

User: ?? ADMM convergence criteria best practices
Claude: I can search public sources for the exact redacted query
        “ADMM convergence criteria best practices”. No workspace files or conversation
        history will be sent. May I send that text to WebSearch and fetch the results?
User: Yes, send that query
Claude: [WebSearch + WebFetch → finds Boyd et al. paper, extracts criteria]
Claude: Boyd recommends ||r|| < ε·max(||Ax||, ||Bz||, ||c||). Use this?
User: Yes
Claude: [Implements]

Best Practices

  • ✅ Always present findings to user before writing code
  • ✅ Use page.fill() for instant text injection instead of keyboard.type()
  • ✅ Ask for fresh approval before every external consultation
  • ✅ Include sources in findings
  • ❌ Don't skip research and write code speculatively
  • ❌ Don't send context, files, or browser data because of a shorthand trigger
  • ❌ Don't alter the user's browser profile or session state

Limitations

  • Requires a user-configured, pinned browser automation connector if browser consultation is used
  • ChatGPT consultation is optional; use ordinary web search when it meets the need
  • GPT response time varies (10-30s typically)
  • Web search quality depends on available sources
  • Does not replace expert domain knowledge — always let user make the final call

Security & Safety Notes

  • Obtain explicit consent for each third-party submission, including the exact redacted text
  • Never access, export, or depend on cookies, saved passwords, or unrelated browser tabs
  • Never submit sensitive credentials, tokens, proprietary code, personal data, or internal URLs
  • Do not install or execute browser tooling from an unpinned package version

Common Pitfalls

Problem Solution
ChatGPT shows login page Let the user log in themselves; do not handle cookies or credentials
The prompt contains sensitive context Redact it or use local reasoning instead
Browser automation is unavailable Use web search or stop and ask the user for a different approved method

Related Skills

  • @systematic-debugging — use when debugging Playwright interactions with ChatGPT
  • @condition-based-waiting — use when waiting for GPT responses in the browser

04 同仓库更多 Skills

Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

testing-best-practices

Laravel test design and review. Use when selecting coverage, naming or structuring tests, choosing assertions or test data, isolating dependencies, testing HTTP or security boundaries, improving suite performance, or reviewing test value. Use framework guidance or search-docs for Pest and PHPUnit syntax.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

tailwindcss-development

Always invoke when the user's message includes 'tailwind' in any form. Also invoke for: building responsive grid layouts (multi-column card grids, product grids), flex/grid page structures (dashboards with sidebars, fixed topbars, mobile-toggle navs), styling UI components (cards, tables, navbars, pricing sections, forms, inputs, badges), adding dark mode variants, fixing spacing or typography, and Tailwind v3/v4 work. The core use case: writing or fixing Tailwind utility classes in HTML templates (Blade, JSX, Vue). Skip for backend PHP logic, database queries, API routes, JavaScript with no HTML/CSS component, CSS file audits, build tool configuration, and vanilla CSS.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

laravel-best-practices

Apply this skill whenever writing, reviewing, or refactoring Laravel PHP code. This includes creating or modifying controllers, models, migrations, form requests, policies, jobs, scheduled commands, service classes, and Eloquent queries. Triggers for N+1 and query performance issues, caching strategies, authorization and security patterns, validation, error handling, queue and job configuration, route definitions, and architectural decisions. Also use for Laravel code reviews and refactoring existing Laravel code to follow best practices. Covers any task involving Laravel backend PHP code patterns.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

infer-conventions

Use this skill to analyze how a Laravel application is actually written and record its conventions as shared rules. Trigger when the user wants to detect, infer, document, or standardize project conventions or coding style, set up or grow `.ai/rules`, resolve mixed or conflicting patterns (e.g. "are we using Form Requests or inline validation?"), or onboard agents and teammates to "how we do things here". Covers: a systematic sweep of ~49 Laravel convention dimensions (validation, models, architecture, testing, frontend, database, console), open-ended house-pattern discovery, conflict reporting, and recording rules scoped to the right paths via the Boost `record-rule` MCP tool. Only run this skill when the user explicitly asks for it; never start a sweep as part of another task. Do not use for one-off code review, enforcing formatting a linter already handles, or editing `.ai/rules` files by hand.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

inertia-react-development

Develops Inertia.js v2 React client-side applications. Activates when creating React pages, forms, or navigation; using <Link>, <Form>, useForm, or router; working with deferred props, prefetching, or polling; or when user mentions React with Inertia, React pages, React forms, or React navigation.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

deploying-to-cloud

Deploys and manages Laravel applications on Laravel Cloud using the `cloud` CLI. Use when the user wants to deploy an app, ship to Cloud, create or manage applications, environments, databases, caches, object storage, queues, domains, instances, background processes, secrets, compute, scheduled tasks, or Laravel Cloud infrastructure; or check billing, usage, or spend. Triggers on deployment, shipping, Cloud management, environment setup, database provisioning, billing or usage queries, the `cloud` CLI, and Laravel Cloud deployment troubleshooting.

★ 1